Table of Content

29 April 2022, Volume 0 Issue 03

Previous Issue    Next Issue

Social Bots Detection Based on Generative Adversarial Networks

LI Yang-yang, YANG Ying-guang

2022, 0(03):  1-6. 

Asbtract ( 281 )   PDF (920KB) ( 160 )  

References | Related Articles | Metrics

Twitter is a social media with hundreds of millions of active users. Nearly 15% of bot accounts are controlled by automated programs. Some of these bot accounts are malicious account that spread malicious information. Although researchers have developed a large number of sophisticated bot account detection methods, they all require prior knowledge of bot accounts which are lack of generalization. In order to solve these problems, this paper proposes to use the discriminator from generative adversarial network for bot account detection. This makes it possible to obtain a good detection model with the examples of real accounts. Experiments on a popular dataset show that the AUC achieves 94% classification effect.

Short-term Load Forecasting Model Based on VMD and MOGOA-LSTM

OUYANG Meng-ke, SHEN Wei-kang, CHENG Hui, SHI Kai

2022, 0(03):  7-12. 

Asbtract ( 173 )   PDF (2495KB) ( 80 )  

References | Related Articles | Metrics

In order to improve the accuracy of short-term load forecasting and reduce the influence of non-stationary signals on model forecasting, this paper proposes a short-term load forecasting model that integrates data mining and multi-objective optimization networks. This method decomposes the power load data into several modal components with different frequencies through variational mode decomposition (VMD) technology, uses the phase space reconstruction (PSR) method to dynamically determine the training of the artificial neural network test ratio and neuron settings, uses the long short-term memory (LSTM) neural network to build models for each component, adds multi-objective grasshopper optimization algorithm (MOGOA) on the basis of LSTM to optimize the internal parameters of the network, and accumulates the predicted values of all component models to realize short-term load forecasting. The simulation results show that, compared with the statistical method and the hybrid model, the proposed model has higher prediction accuracy and stronger generalization ability in short-term prediction.

End to End Voiceprint Recognition Based on Nonlinear Stacked Bidirectional Network

WANG Zhi-yue, CUI Lin,

2022, 0(03):  13-17. 

Asbtract ( 170 )   PDF (831KB) ( 90 )  

References | Related Articles | Metrics

The traditional voiceprint recognition method is cumbersome and has a low recognition rate. The neural network used in the existing deep learning method is not specific to the speech signal, resulting in insufficient recognition accuracy. To solve the above problems, this paper proposes an end-to-end voiceprint recognition method based on nonlinear stacked bidirectional LSTM. Firstly, the Fbank features are extracted from the original voice files for the input of the network model. Then, in view of the continuous and strong relevance of the voice signal, a bidirectional long and short-term memory network is constructed to process the voice data to extract deep features. In order to further enhance the nonlinear expression ability of the network, stacking multi-layer bidirectional LSTM layer and multi-layer nonlinear layer are used to extract the deeper abstract features of the speech signal. Finally, the SGD optimizer is used to optimize the training mode. The experimental results show that the proposed method can make full use of the characteristics of the speech sequence signal and has strong time series comprehensiveness and nonlinear expression ability. The constructed model has strong integrity and better recognition effect than GRU and LSTM models.

Power Demand Forecast of Different Types in Multiple Regions Based on Extreme Gradient Boosting

ZHANG Su-ning, WANG Fang, ZHU Yan, JING Dong-sheng

2022, 0(03):  18-22. 

Asbtract ( 100 )   PDF (925KB) ( 63 )  

References | Related Articles | Metrics

The forecasting of heterogeneous power demands in multiple regions not only ensures the stability of the power supply, but also reasonably distributes heterogeneous power resources produced nationwide. However, existing approaches mainly forecast single time series for single region, which cannot meet the power demand forecasting requirements for complex situations in the energy Internet. To solve the problem, an algorithm based on extreme gradient boosting is designed, which is able to predict the demand of multi-category power over different regions. The proposed algorithm improves the boosting tree method and effectively prevents over fitting. Meanwhile, it also improves the training efficiency by supporting distributed parallelization. Compared with other methods, the proposed method is less stringent on the total amount of training samples and characteristic data types and can be used for multi-time series forecasting. The experimental results show that the proposed algorithm can predict the different types of power demand in different regions quickly and accurately within the acceptable range of error.

Vehicle Re-identification Method Based on Non-local Attention and Local Features

WAN Dong-hou, ZHANG De-xian, DENG Miao-lei,

2022, 0(03):  23-29. 

Asbtract ( 179 )   PDF (3323KB) ( 174 )  

References | Related Articles | Metrics

Vehicle re-identification refers to re-identifying the same vehicle from different cameras. The result of vehicle re-identification is easily affected by other factors such as vehicle angle and illumination, which is a very challenging task. Many vehicle re-identification methods pay too much attention to the global features of the vehicle, but ignore the local resolution features of the vehicle image, which result in the problem of low accuracy of vehicle re-recognition. To solve this problem, this paper proposes a vehicle re-identification method integrating non-local attention and multi-scale features. The attention mechanism is used to obtain vehicle salient features and integrate multi-scale features, so as to improve the retrieval accuracy of vehicle re-identification. Firstly, the backbone feature extraction network and attention module are used to obtain the significant fine-grained features of vehicles. Then, the feature is divided into multiple branches for metric learning. The local and global features of vehicles are learned respectively, and the global features and fine-grained local features are fused to construct the features of vehicle re-identification. Finally, this method is used to extract the characteristics of different vehicles and calculate the similarity of different vehicles and judge whether they have the same identity. The experimental results show that the vehicle re-identification algorithm using attention mechanism and local features has higher accuracy.

Secret Content Auditing Scheme in Internet of Things Scenes Based on Blockchain

ZHANG Tian-xi, WANG Li-peng, ZHOU Chun-tian, YANG Yan-yan, LI Xiao-chong

2022, 0(03):  30-36. 

Asbtract ( 153 )   PDF (1433KB) ( 73 )  

References | Related Articles | Metrics

Low-power wide area network has the advantages of low power consumption and long distance, which has become a research hotspot. However, the system is not completely reliable in terms of efficiency and privacy due to the limited security capability of sensor nodes, easy aging and malicious nodes in low-power wide area network. Aiming at the above problems, this paper analyzes Lora’s media access control layer and proposes a set of ciphertext data audit algorithm. The efficiency of ciphertext data processing is accelerated by Bloom filter. In addition, the homomorphic encryption technology is used to realize effective ciphertext data range query function on the premise of data protection. Finally, the Blockchain technology is used to realize the traceability of communication data, so as to facilitate the follow-up of problem nodes. The analysis results show that the new scheme has high security performance and the communication efficiency can reach O(logn).

Source Traceability Method for Power Consumption Data Leakage Based on Watermark and Attribute Screening

SHAN Chao, ZOU Yun-feng

2022, 0(03):  37-42. 

Asbtract ( 139 )   PDF (934KB) ( 63 )  

References | Related Articles | Metrics

Electricity consumption data involves customer’s privacy, and during the distribution and sharing process, it has the risk of unauthorized outgoing. Digital watermarking is an effective technology to realize leak traceability and hold them accountable. The implantation of watermarks will cause data offset and affect the usability of data analysis and the traceability effect is not good enough when some part of the data is leaked. In view of the above problems, this paper proposes an electricity data traceability algorithm WRTA based on sub-watermark and attribute filtering, which uses information gain rate and Gini coefficient to measure the influence of attributes on maintaining classification availability, and randomly selects non-important attributes through key and primary key to construct sub-watermarks. The algorithm can provide the availability and security of data analysis and realize the traceability of partial data leakage.

Adaptive Encryption Method for Private Database Based on Branch Obfuscation Algorithm

ZHANG Li, LUO Chun-shan, XIE Wei-yuan, LI Bei-bei

2022, 0(03):  43-47. 

Asbtract ( 150 )   PDF (920KB) ( 61 )  

References | Related Articles | Metrics

In order to effectively reduce the possibility of data leakage in private database and further improve data security, an adaptive encryption method for private database based on branch obfuscation algorithm is designed in this paper. On the basis of collecting the request contents of the application program to the database, the characteristics of the data in the privacy database are classified by using the branch obfuscation algorithm, and the confusion points in the database are classified by selecting the Naive Bayes classifier, so as to determine the field types of the data in the database. Based on this, data encryption algorithms are selected for different field types. Among them, the sequence-preserving encryption algorithm is used for the numerical data, the equivalent encryption algorithm is used for the equivalent comparison data, and the word segmentation assisted index encryption algorithm is used for the text data. While encrypting numeric, time and character fields, the partial order and searchable properties are preserved. The experimental results show that the proposed method can effectively encrypt the data of different field types in the privacy database, and the encryption requires less time and has high security.

Text Clustering Algorithm Based on RoBERTa-WWM and HDBSCAN

LIU kun, ZENG Xi, QIU Zi-heng, CHEN Zhou-guo,

2022, 0(03):  48-52. 

Asbtract ( 421 )   PDF (1092KB) ( 72 )  

References | Related Articles | Metrics

In the big data environment, obtaining hot topics from massive Internet data is the basis for studying public opinion and sentiments in the current Internet. Among them, text clustering is one of the most common methods to get hot topics, which can be divided into two steps: text vectorization representation and clustering. However, in the task of vectorized text representation, the traditional text representation model cannot accurately represent the contextual information of texts such as news and posts. In the clustering task, the K-Means algorithm and DBSCAN algorithm are most commonly used, but their clustering method is not consistent with the actual distribution of topic data, which makes the existing text clustering algorithms very poorly applied in the actual Internet environment. Therefore, this paper proposes a text clustering algorithm based on RoBERTa-WWM and HDBSCAN according to the data distribution of topics in the Internet. Firstly, the pre-trained language model RoBERTa-WWM is used to obtain the text vector of each text. Secondly, the t-SNE algorithm is used to reduce the dimension of the high-dimensional text vector. Finally, the HDBSCAN algorithm based on hierarchical density clustering algorithm is used to cluster the low-dimensional text vector. The experimental results show that compared with the existing text clustering algorithms, the proposed algorithm has a great improvement in the clustering effect on data sets that contain noisy data and are unevenly distributed.

Implicit Tag Collaborative Filtering Recommendation Algorithm Based on LDA

WEN Yong-jun, HE Huan-jing, TANG Li-jun,

2022, 0(03):  53-58. 

Asbtract ( 116 )   PDF (1134KB) ( 75 )  

References | Related Articles | Metrics

The fixed tag collaborative filtering recommendation algorithm does not fully consider the diversity of tag factors, and mainly relies on manual tagging, which is not scalable and has many subjective factors. In this paper, based on the fixed tag collaborative filtering recommendation algorithm, an implicit tag collaborative filtering recommendation algorithm is proposed from the perspective of user preferences. This algorithm uses LDA topic model to generate implicit tags of item text, and obtains item-tag feature weights. The number of tags is selected according to the requirements of algorithm performance optimization, and the user’s preference matrix for tags is obtained by combining the item-tag matrix with the user scoring matrix. Finally, the recommendation is generated by collaborative filtering algorithm. The experimental results show that the user-based LDA tag collaborative filtering algorithm proposed in this paper alleviates the problem of data sparsity, and greatly improves the recall rate, accuracy and F1 value of item recommendation.

Skeleton Extraction Algorithm Based on Heat Method

SU Chen-yao, LIU Xiang-yang

2022, 0(03):  59-63. 

Asbtract ( 153 )   PDF (1939KB) ( 151 )  

References | Related Articles | Metrics

Skeleton extraction is an important branch of shape representation and has a wide range of applications in the image compression, pattern recognition and shape matching. This paper proposes a skeleton extraction algorithm based on heat method. The algorithm first constructs triangle meshes for the target region, finds the direction of increasing distance by solving the heat conduction equation, and uses the Poisson equation to restore the geodesic distance. Then the endpoints of the skeletons are determined by using the voting method, and the continuous skeleton lines are obtained by the paths backtracking. Finally, the skeletons of loops are detected and extracted by clustering the end points. The algorithm is robust, accurate and simple to implement since it only needs to solve a pair of standard sparse linear systems. In addition, some information in the pre-calculated can be reused, which reduce memory footprint and time consumption. The experimental results show that the algorithm can extract the skeleton of target shape accurately.

Underwater Crack Segmentation Algorithm Based on Polarization Characteristic Map

ZOU Jie, SONG Ke, WANG Zhang-fan, HOU Yi-xing, ZHANG Xue-wu

2022, 0(03):  64-69. 

Asbtract ( 127 )   PDF (3144KB) ( 70 )  

References | Related Articles | Metrics

Due to water erosion, natural aging and other problems, underwater structures such as reservoirs have surface damage defects during long-term operation. The depressions formed by attachments such as moss and shells on the surface of underwater structures are highly similar to cracks in terms of color and texture. When the depressions are close to cracks, it is easy to cause misjudgment in crack image segmentation. In order to solve this problem, the polarization information is introduced in this paper, and a crack image segmentation algorithm based on polarization degree feature map is proposed. The feature map is extracted by superpixel block, the region growth rule is set by polarization degree, and the region growth is combined based on seed region. Finally, the crack region is determined according to the image depth map. The experimental results show that this algorithm can effectively reduce the segmentation misjudgment, and the specificity and accuracy index of crack segmentation results can reach more than 0.9, which is much higher than other algorithms.

Spatio-temporal Fall Event Detection Algorithm Based on Attention Mechanism Subnetwork

XIE Hui, SHI Hou-qing, QI Yu-xiao, CHEN Rui, TONG Ying

2022, 0(03):  70-75. 

Asbtract ( 160 )   PDF (4720KB) ( 82 )  

References | Related Articles | Metrics

In recent years, it has become increasingly important to automatically detect falls and provide feedback in emergency situations. In complex scenes, aiming at the problem of the degradation of detection performance due to changes in illumination, occlusion and scale changes, a real-time and robust fall detection algorithm is proposed. Firstly, the YOLO v3 target detection module is used to complete pedestrian detection. Then, after extracting the deep features of each tracked bounding box in the tracking module, data enhancement and re-detection techniques are used to improve the detection accuracy under light changes, and the attention mechanism subnetwork is introduced to deal with the detection of obscured targets. Finally, the final fall judgment module is used to judge the pedestrian posture and complete real-time fall detection and alarm. The experimental results on the Cityperson data set, Montreal fall data set and self-built data set show that the detection accuracy of the pedestrian detection algorithm reaches 87.05%, the detection accuracy of the fall algorithm reaches 98.55%, and the delay is within 120 ms. Good performance can still be obtained under the influence of occlusion.

Rotation Calibration Method of Concentric Circles for Line Scan Camera

TIAN Zhong, WU Shi-qian

2022, 0(03):  76-81. 

Asbtract ( 294 )   PDF (1902KB) ( 65 )  

References | Related Articles | Metrics

The calibration process of traditional line scan cameras is complex and requires high accuracy of calibration objects, it is difficult to ensure the positioning accuracy of defects. This paper proposes a ring rotation calibration method for line scan cameras to improve the positioning accuracy of defects. This method designs a new type of circular calibration board. On the basic of static calibration, the coordinates of the intersection of the camera’s line of sight and the circle are collected by a rotating line camera to obtain the rotation angle and multiple sets of calibration points to establish the imaging model and radial distortion model of the line camera. The internal parameters and distortion parameters of the camera are solved by nonlinear optimization of the global overall function, and the impact of different camera rotation angles on the calibration accuracy are analyzed. The experimental results show that when θ≤20°, the calibration accuracy of this method is within 0.35 pixel, which meets the positioning requirements of actual detection. The PCB defect detection is well verified.

Defect Detection of Automobile Parts Based on ECA-SSD Model

JIN Wen-qian, PENG Lu-lu, ZHU Yuan-yuan, WANG Xiao-mei

2022, 0(03):  82-90. 

Asbtract ( 173 )   PDF (6873KB) ( 60 )  

References | Related Articles | Metrics

Automobile parts have great influence on the appearance, performance and safety of automobile. Due to the large number of automobile parts, small volume and high accuracy requirements, there are certain requirements for the accuracy and speed of parts detection. Using image processing technology, based on SSD model, the VGG module is replaced by deep separable convolution and linear bottleneck inverse residual structure. An effective attention mechanism ECA module is introduced to avoid dimensionality reduction. At the same time the computational complexity of the model parameters is reduced, and the channel is increased to improve the accuracy of the model. And this paper focuses on the image target, ignores the interference of the background to achieve fast and accurate defect detection of automotive parts. In addition, the proposed model in this paper is used to detect the outer wall defects of automobile parts provided by SAIC. The experimental results show that the size of the model is only 15.9 MB, the mAPis 94.64%, and the detection time of each image is 0.013 s, which meets the requirements of speed and accuracy in the automotive industry. Compared with other target detection algorithms such as VGG-SSD、MobileNetv2-SSD and YOLO v3, the detection accuracy, speed and size of the proposed model are improved.

Grid Generation Automation Algorithm of Multi-scale Component Structures Based on Feature Recognition

CHEN Zhong-jie, TIAN Jian-hui, HU Guang-chu, DING Feng, GUO Zhao, HAN Xing-ben

2022, 0(03):  91-97. 

Asbtract ( 95 )   PDF (4255KB) ( 62 )  

References | Related Articles | Metrics

With the increasing complexity of the problems to be solved in the field of structural mechanics, the high-quality mesh generation of multi-scale component structures plays an important role in the computational accuracy of the numerical simulation. This paper proposes a method for grid generation automation based on feature recognition. In this method, the geometric features of different scale of complex multi-scale component structures are identified, the grid size values of related regions are set according to their different scale sizes. The delaunay triangulation algorithm or advancing front technique are used to generate grid elements，which can reflect the geometric characteristics of different scale, and then the surrounding areas of small scale are encrypted. Finally, the meshes of different scales are connected by geometric exponential control function to form the whole meshing model of complex multi-scale component sructures. The test results of two geometric models show that the overall grid quality generated by this method is good, the mesh transition of different scales is reasonable transition, and the automation degree is high.

Ant Colony Algorithm Combining Neighborhood Search Strategy for Vehicle Routing Problem with Time Window

ZHANG Xiong, PAN Da-zhi,

2022, 0(03):  98-102. 

Asbtract ( 114 )   PDF (841KB) ( 66 )  

References | Related Articles | Metrics

For solving the vehicle routing problem with time windows, this paper proposes an improved ant colony algorithm that integrates neighborhood search strategies. In view of the time window characteristics, the waiting time is added to the state transition rule of the ant colony algorithm. In order to improve the local optimization ability of the algorithm, a variety of node for deletion operations and insertion operations are designed to search the neighborhood of the obtained path. Finally, the improved algorithm is tested using the Solomon standard example, and compared with the currently known optimal solution. The experimental results show that the improved ant colony algorithm has good applicability to the vehicle routing problem with time windows.





Enhance Flexibility of Graph Convolutional Filter Based on Graph Filter Framework

XU Xin-qiang, HE Peng,

2022, 0(03):  103-110. 

Asbtract ( 69 )   PDF (1336KB) ( 63 )  

References | Related Articles | Metrics

Graph convolutional networks learn the convolution kernels through feature propagation to achieve graph convolution. Its core lies in the construction of the convolution operator. When applied to specific graph data, the applications of convolution kernels are often limited due to the difference of scenarios. This paper addresses the convolution kernels from a graph filtering perspective. Under the graph filtering framework, the data features of nodes are regarded as graph signals, and the smooth signals are processed by low-pass filter. The extracted smooth graph signals are placed on topographic map for convolution in spectral domain. In this process, the local graph structure information will be integrated into the similarity representation of the nodes to complete the graph embedding learning. In order to improve the flexibility of the graphics filter and achieve more detailed design, this paper naturally extends the original model and introduces a new balance parameter, which can easily control the smoothness of the filter to meet the filtering requirements of various application scenarios without increasing the number of trainable weights of the neural network, and its mechanism is to control the horizontal displacement of the frequency response function. By setting a variety of parameter values on three citation networks and a knowledge graph to perform the task of graph embedding learning, this paper verifies the effectiveness of introducing the balance parameter and proposes a more comprehensive view from the perspective of graph partition.

Sparse Signal Reconstruction of Backtracking Generalized Orthogonal Matching Pursuit Algorithm Based on Secondary Screening

ZHANG Lian-na, ZHANG Hui-ping, LI Rong-peng, SONG Xue-li

2022, 0(03):  111-115. 

Asbtract ( 117 )   PDF (836KB) ( 65 )  

References | Related Articles | Metrics

Compressed sensing is a new theory of signal sampling and reconstruction. Efficient signal reconstruction algorithm is the pivot of compressed sensing from theory to practical application. To reconstruct the original sparse signal more accurately，a backtracking generalized orthogonal matching pursuit algorithm based on secondary screening is proposed. Firstly, a large number of related atoms are selected to improve their utilization rate by using inner product matching criterion. Secondly, the generalized Jaccard coefficient criterion is used for the selected atoms to obtain the most matched atoms and optimize the atom selection method. The experimental results show that when the signal is reconstructed under different sparseness and observed values, the proposed algorithm has greater advantages in terms of reconstruction error and  success rate compared with backtracking generalized orthogonal matching pursuit algorithm, orthogonal matching pursuit algorithm and subspace pursuit algorithm.

Review on Fault Diagnosis Technology of Transformer

LIN Fan-qin, LI Ming-ming, GUO Hong

2022, 0(03):  116-126. 

Asbtract ( 761 )   PDF (1589KB) ( 127 )  

References | Related Articles | Metrics

Transformer equipment plays an essential role in the power system. Its healthy and stable operation is related to realizing the function of power distribution voltage conversion, and fault diagnosis technology can escort for the transformer’s regular operation. This paper summarizes the research status of transformer fault diagnosis technology at home and abroad, analyzes the development process of transformer fault diagnosis, compares the merits of different diagnosis methods, and analyzes the traditional method of extracting transformer fault data——dissolved gas extraction method and sound signal. Finally, the research focus and development trend of transformer fault diagnosis in the future are put forward to provide some reference for transformer fault diagnosis.